Related Experiment Video
Updated: Sep 7, 2026
Multi-modal Pulmonary Imaging: Using Complementary Information from CT and Hyperpolarized 129Xe MRI to Evaluate Lung Structure-Function
Published on: April 12, 2024
Ai-assisted compressed sensing with deep learning reconstruction for accelerated rectal MRI: A prospective
Yuedi Ma1, Dongqiu Shan2, Junhui Yuan1
1Department of Medical Imaging, The Affiliated Cancer Hospital of Zhengzhou University & Henan Cancer Hospital, Zhengzhou 450008, Henan Province, China.
Objectives:
This study aimed to compare image quality and diagnostic performance between artificial intelligence-assisted compressed sensing (ACS) images reconstructed using deep learning reconstruction (ACS-DLR) and conventional parallel imaging (PI) images in rectal cancer MRI.
Methods:
107 patients with biopsy-proven rectal cancer were included. MRI included conventional PI and ACS acquisitions, with the ACS raw data reconstructed at three deep learning reconstruction strength levels (ACS-L, ACS-M, and ACS-H). Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were compared across four image sets using the Friedman test. Subjective image quality was assessed using a 5-point Likert scale for overall image quality, noise, artefact, and edge sharpness. Interobserver agreement for objective metrics was measured by ICC, and for subjective metrics by Cohen's kappa. Diagnostic performance was evaluated using postoperative histopathology, including T stage, N stage, extramural venous invasion (EMVI), and mesorectal fascia (MRF) involvement.
Results:
ACS reduced acquisition time by 50 % (from 3 min 20 s to 1 min 40 s). Lesion SNR did not differ significantly among the four image sets (P > 0.05), but ACS-H showed the highest muscle SNR. CNR showed significant differences in selected pairwise comparisons. ACS-H achieved the highest subjective scores for overall image quality, noise reduction, and lesion edge sharpness. In the surgical subcohort, ACS-H improved T staging accuracy (P = 0.010; P = 0.018), MRF involvement assessment (P = 0.004; P = 0.012), and EMVI sensitivity (P = 0.039; P = 0.041). N staging accuracy was not significantly different (P = 0.521; P = 0.841).
Conclusion:
ACS reduced acquisition time, while ACS-DLR improved subjective image quality. ACS-H improved T-stage and MRF assessment and increased EMVI sensitivity, whereas N-stage accuracy did not improve significantly.